Deeply understand the problem. We spend a lot of time with our customers to understand their processes and pain points. Often times, latent needs are not communicated but obvious through observation. Our product has been informed by working alongside clients and looking for common problems.
As a part of our series about cutting edge technological breakthroughs, I had the pleasure of interviewingYashar Behzadi.
Yashar Behzadi is an experienced entrepreneur who has built transformative businesses in AI, medical technology, and IoT markets. Now the CEO at Synthesis AI, he has spent the last 14 years in Silicon Valley building and scaling data-centric technology companies. His work at Proteus Digital Health was recognized by Wired as one of the top 10 technological breakthroughs of 2008 and as a Technology Pioneer by the World Economic Forum. Yashar has over 30 patents and patents pending and a Ph.D. in Bioengineering from UCSD.
Thank you so much for doing this with us! Can you tell us a story about what brought you to this specific career path?
My parents are both architects and builders, and as a child, I always marveled at how an empty dirt lot could be transformed into a beautiful house. The process of building a house is one of vision, establishing a foundation, diligently planning and minding dependencies, and carefully attending to details. This is analogous to building a company and product. This early exposure catalyzed my passion for coming up with white-space ideas and reducing them into tangible products.
I have been lucky enough to work on cutting edge products, including DNA fingerprinting, digital medicines, and advanced computer vision systems. I am currently excited about reimaging how AI systems are built and creating a new paradigm that solves the ‘data problem’ in AI. I feel that we can have an impact at scale, and by creating a new paradigm for AI development, we will enable a new class of more capable products.
Can you share the most interesting story that happened to you since you began your career?
I was fortunate enough to work on some incredible technology at a pioneering digital health company. Our ambitious goal was to create an ingestible sensor that could be embedded in all pharmaceuticals to transform medications into smart digital products that could provide detailed information about ingestion events and biometrics. It was audacious and technically very difficult. I was responsible for architecting a communication system between this tiny, ingestible sensor to a patient’s smartphone and doctor’s dashboard. After months of simulation and algorithm development, I was able to see the system work end-to-end in a live demo for a large pharmaceutical company. The company later invested millions of dollars into the company. It was an amazing experience to see work done in simulation work in reality, even with the tremendous complexity involved.
Can you tell us about the Cutting edge technological breakthroughs that you are working on? How do you think that will help people?
Synthesis AI is pioneering the use of simulation and synthetic data to build more capable computer vision AI. Phones, homes, cars, and industrial robots heavily leverage vision systems to understand the world and make decisions. These systems are currently built using human-curated and labeled data to teach the AI systems how to interpret images. This process is expensive & slow and limited to what humans can effectively label. Historically, there has also been little regard for consumer privacy, and these concerns are amplified as systems increasingly use human facial data in our home environments.
Our approach is to create photorealistic digital worlds in which complex image data can be synthesized. Since we generate the data, we know everything about the scenes, including never before available information about the 3D location of objects and their complex interactions with one another and the environment. We are able to create millions of perfectly-labeled images on demand. Acquiring and labeling this amount of data using current approaches would take months, if not years. This new paradigm will enable a 100x improvement in efficiency and cost and drive a new class of more capable models.
How do you think this might change the world?
AI systems are ubiquitous but contain inherent biases that may disproportionately impact groups of people. Synthetic data approaches will lead to the development of more ethical and less biased AI systems.
Human bias is inherent in the current human labeling paradigm. Datasets are often misbalanced with certain classes of data either over or under-represented. When building human-centric systems, this leads to gender, ethnicity, and age biases. Bias in AI systems has been a major concern of both consumers and regulators. In contrast, generated training data by design is properly balanced and lacks human biases.
In addition to concerns around bias, privacy is a growing concern. Technology companies have shown little regard for privacy in building AI systems, and new synthetic data approaches solve this issue at the data level.
Keeping “Black Mirror” in mind can you see any potential drawbacks about this technology that people should think more deeply about?
We provide synthetic data to enable the development of AI models. We are careful to work with companies with a clear ethical position on using AI in their product systems. As with any AI product, there are alternative use-cases (e.g., government surveillance, military applications) that have consequences for individuals. We explicitly do not support these use-cases and focus on driving value to consumers in an unbiased and privacy-minded manner. Unfortunately, there are competing companies in the space that have taken a different stance.
Was there a “tipping point” that led you to this breakthrough? Can you tell us that story?
We were working with a major handset manufacturer to support their face unlock capability. Traditionally, companies would bring human actors to a studio and capture images across a variety of use-cases. This process is further complicated as handset manufacturers consider various optic systems. To properly assess performance, data would have to be captured with various hardware. The data would then have to be labeled by humans. The process of building hardware, acquiring and labeling images is tedious, time-consuming, and expensive. We synthetically created 10’s of thousands of unique 3D humans to support this use-case and generated millions of pixel-perfect labeled data. We were also able to capture many use-cases that were previously difficult to capture, such as variability of the environment, lighting conditions, and appearance changes related to glasses, masks, hairstyles, make-up, etc. We ultimately demonstrated that synthetic data was 100x more efficient to leverage, leading to higher performing and more generalized models. This was a significant milestone as it demonstrated that our approach could improve models that took years and millions of dollars to develop in a few days.
What do you need to lead this technology to widespread adoption?
It is inevitable that simulation and synthetic data will be used to develop computer vision AI. To reach widespread adoption, we need to continue to build out 3D models to represent more of the real-world and create scalable cloud-based systems to make the simulation platform available on-demand across a broad set of use-cases. We also plan to integrate our offering into existing machine learning and cloud platforms to make the overall development experience seamless for machine learning developers.
What have you been doing to publicize this idea? Have you been using any innovative marketing strategies?
We have been focused on rigorously validating our technology and engaging the AI ecosystem. We have been lucky enough to be showcased by AWS and co-publish with Google. We also have published the most comprehensive technology review of the space and have a popular technical blog. We will be releasing a book on Synthetic Data this year to solidify our thought-leadership further. In the coming months, we will begin to target enterprise companies and specific use-cases. Our current customers and partners will co-publish with us, and we are excited to announce some of the great work we have been doing in close collaboration with these leading companies.
None of us are able to achieve success without some help along the way. Is there a particular person who you are grateful towards who helped get you to where you are? Can you share a story about that?
I am personally very thankful to my parents for their inspiration and my spouse for her support. Running a start-up is intense and requires a lot of personal sacrifices, and without strong family support, it’s a difficult journey to undertake.
I am also incredibly thankful for our team. I marvel at their ability to solve complex problems and work together to deliver for our customers.
How have you used your success to bring goodness to the world?
I am hopeful that the products we build will lead to more ethical AI. In a rush to commercialize AI, many companies, intentionally or unintentionally, valued profit over building inclusive, ethical AI systems. By creating a 100x more efficient paradigm to build AI models, companies will no longer have to trade-off economics for ethics.
What are your “5 Things I Wish Someone Told Me Before I Started” and why.
- Deeply understand the problem. We spend a lot of time with our customers to understand their processes and pain points. Often times, latent needs are not communicated but obvious through observation. Our product has been informed by working alongside clients and looking for common problems.
- Delight customers. I recall one customer who was initially unhappy with the data we provided. Although we delivered against the outlined statement of work, the customer later realized that the data they needed was not what had been specified. At a significant cost to us, we iterated until we could drive a key model performance metric for the client. The client is now our largest customer.
- Culture matters. The tone and values of the founders establish the cultural base of the company. Earlier in my career, I was brought into companies as an executive to help with product and technology pivots. The company was often not held back by the team’s capability, product, or technology, but rather by the culture.
- Enable your team. I believe in aligning authority, autonomy, and accountability with my senior team members. By providing context and enabling the team to do their best work, our team is happier and more productive.
- Ethics are the foundation. I have said no to certain customers, investors, and partnering opportunities because they do not align with our company’s focus on building ethical AI systems. Near-term profit is not worth compromising the ethical backbone of your company.
You are a person of great influence. If you could inspire a movement that would bring the most amount of good to the most amount of people, what would that be? You never know what your idea can trigger. 🙂
Traditionally large companies invested years and millions of dollars in building ‘data moats’ to create a competitive advantage. Synthetic data has the power to disrupt the space as it enables small, agile start-ups to build world-class AI models efficiently and cost-effectively. I hope this results in many more people across the globe contributing to and benefitting from benevolent AI.
Can you please give us your favorite “Life Lesson Quote”? Can you share how that was relevant to you in your life?
“The busy get lucky” was something a friend of mine told me when I started my first company. Many early-stage opportunities come from meeting the right people at the right time. By keeping busy, I have met future team members, customers, and investors who have been transformative in driving business success.
Some very well known VCs read this column. If you had 60 seconds to make a pitch to a VC, what would you say? He or she might just see this if we tag them 🙂
By 2022 there will be 45B connected cameras in the world, all driven by computer vision AI.
However, companies are limited by the availability of sufficiently diverse and accurately human-labeled datasets. More fundamentally, humans are unable to label many key attributes required for emerging applications.
Synthesis AI, an SF-based technology company, is pioneering the use of synthetic data to build more capable computer vision models. The company’s data generation platform is a 100x solution for computer vision and will lead to the development of more capable computer vision models. Showcased by AWS and Google, Synthesis AI is ushering in a new paradigm for AI model development.
How can our readers follow you on social media?
Thank you so much for joining us. This was very inspirational.